Spiking Networks as Non Local Cellular Automata
In this paper, we propose a novel regularization method for spiking neural networks. We note the similarities of a spiking network to a non local cellular automaton and derive a rule for its connections that must be learnt. This rule results in the activations of the network forming a grammar. We also view each state of the spiking network as a node on a hyper graph and show that we can generalise by simply sending the never before seen acivations presented at a novel state or situation, to the next activations that the grammars rule would have them move to. We also demonstrate that hierarchical planning may be achieved by the particular grammar we choose.
https://www.researchgate.net/publication/384809319_Spiking_Networks_as_Non_Local_Cellular_Automata
1 Like
If anyone would like to be a coauthor on a conference paper based on this idea please contact me.
2 Likes
https://www.researchgate.net/publication/385214815_An_Algorithm_to_Optimize_the_Routing_of_Neurons_in_an_Artificial_Spiking_Neural_Network
We describe an algorithm that determines the optimal connectivity of a spiking neural network such that it can reach the lowest possible loss for that number of neurons. If the spiking network was as big as the human brain, then a local minimum would have it connected just like a typical brain is connected after the algorithm is done. Giving it all of its functionality to the extent that artificial spiking neurons mimic biological ones.
im working on a proof that hinges on the way this process decouples the functions compositions and modules
1 Like
You have to have some code, there could be value in this concept but since the feedback seems pretty low, is your job to figure out implementing it on an actual problem.
2 Likes
Im currently working on code for a related area. Thanks for the advice!
1 Like